The Genetics of Hemoglobin A₂ Regulation in Sickle Cell Anemia

Abstract

Hemoglobin A₂, a tetramer of α- and δ-globin chains, comprises less than 3% of total hemoglobin in normal adults. In northern Europeans, single nucleotide polymorphisms (SNPs) in the HBS1L-MYB locus on chromosome 6q and the HBB cluster on chromosome 11p were associated with HbA₂ levels. We examined the genetic basis of HbA₂ variability in sickle cell anemia using genome-wide association studies (GWAS). HbA₂ levels were associated with SNPs in the HBS1L-MYB interval that affect erythropoiesis and HbF expression and SNPs in BCL11A that regulate the γ-globin genes. These effects are mediated by the association of these loci with γ-globin gene expression and fetal hemoglobin (HbF) levels. The association of polymorphisms downstream of the β-globin gene (HBB) cluster on chromosome 11 with HbA₂ was not mediated by HbF. In sickle cell anemia, levels of HbA₂ appear to be modulated by trans-acting genes that affect HBG expression and perhaps also elements within the β-globin gene cluster. HbA₂ is expressed pancellularly and can inhibit HbS polymerization. It remains to be seen if genetic regulators of HbA₂ can be exploited for therapeutic purposes.

Introduction

Hemoglobin A₂ (HbA₂: α₂δ₂), a tetramer of α- and δ-globin chains, forms less than 3% of total hemoglobin in normal adults.[1] It has no known physiological function, but elevated levels are associated with β-thalassemia trait. Its levels are heritable. Genetic variation explains 42% of total HbA₂ variability using a variance components model.[2] The genetic basis of HbA₂ regulation has been studied in northern Europeans where single nucleotide polymorphisms (SNPs) in the HBS1L-MYB locus on chromosome 6q and the HBB cluster on chromosome 11p were associated with HbA₂. This effect is presumably a result of modifying the kinetics of erythropoiesis and transcription within the HBB gene cluster.[2]

HbA₂ can inhibit the polymerization of sickle hemoglobin (HbS) and it has a pancellular distribution.[3-6] High HbA₂ might therefore be of benefit in sickle cell anemia.[7] Polymorphisms of the HBS1L-MYB interval that are associated with hematopoiesis and fetal hemoglobin (HbF) concentration are differentially distributed among populations,[8] and higher than normal HbF levels are characteristic of sickle cell anemia.[9] Stress erythropoiesis might also be partly responsible for increased HbF levels. A reciprocal relationship between HbA₂ and HbF levels is present in acquired disorders where HbF levels are increased [10] and HbA₂ levels are lowest in cells with increased HbF and in individuals with high HbF.[11, 12] Menzel found that females with more F-cells had lower HbA₂ than males with fewer F-cells.[2] To confirm and extend the observations on the genetic regulation of HbA₂ we examined the genetic basis of HbA₂ variability in sickle cell anemia.

Materials & methods

Study subjects

The discovery cohort included 618 unrelated African American subjects from the Cooperative Study of Sickle Cell Disease (CSSCD; NCT00005277).[13] To replicate the associations with p<1E-5 we used 128 African American patients from the Pulmonary Hypertension and Sickle Cell Disease with Sildenafil Therapy (Walk-PHaSST;NCT00492531)[14], 45 African American cases from the Pulmonary Hypertension and the Hypoxic Response in Sickle Cell Disease (PUSH; NCT 00495638) [15], and 580 Chinese subjects from the Hong Kong β-thalassemia trait study (Table 1).[16] The studies were approved by the Institutional Review Boards of all participating sites.

Table 1.

Replication of results from (Menzel et al., 2013) [2] in the CSSCD cohort.

CHR		SNP	BP	Menzel P	P adj	CSSCD Effect*	P*	Effect adj**	P adj**	Effect HbF***	P HbF***	Effect Mediation+	P Mediation+
Chromosome 6q
	6	rs1547247	135432529	3.02E-08	1.39E-05	−0.17	2.17E-02	−0.12	1.07E-01	1.72	2.22E-03	− 0.05	6.30E-03
	6	rs7775698	135460328	2.51E-09	3.30E-05	−0.08	4.54E-02	−0.06	1.54E-01	0.81	1.02E-02	− 0.03	1.78E-02
	6	rs9399137	135460711	5.19E-09	4.50E-05	−0.17	1.98E-02	−0.08	2.66E-01	2.97	9.22E-08	− 0.09	8.16E-05
	6	rs4895441	135468266	5.72E-09	5.50E-05	−0.09	1.25E-01	−0.03	5.46E-01	1.72	8.06E-05	− 0.05	9.12E-04
	6	rs9376092	135468837	8.69E-09	7.16E-05	−0.03	4.71E-01	−0.02	7.07E-01	0.46	1.42E-01	− 0.01	1.52E-01
	6	rs9494145	135474245	1.03E-08	1.42E-05	−0.22	9.81E-04	−0.13	4.10E-02	2.78	1.81E-08	−0.08	6.33E-05
Chromosome 11p
	11	rs11036212	5178401	1.45E-09	1.03E-08	−0.08	2.26E-02	−0.07	4.50E-02	0.37	1.67E-01	− 0.01	1.77E-01
	11	rs7950726	5182023	1.23E-11	1.99E-10
	11	rs12787404	5182342	6.82E-12	1.01E-10
	11	rs10837582	5185284	3.12E-08	7.23E-08	0.03	6.62E-01	0.02	7.23E-01	−0.21	7.01E-01	0.01	7.02E-01
	11	rs12793110	5188141	5.11E-12	7.73E-11
	11	rs10837628	5200980	1.53E-09	2.41E-09	0.03	6.60E-01	0.02	7.46E-01	−0.28	6.05E-01	0.01	6.06E-01
	11	rs11036364	5205580	2.80E-09	2.59E-08	0.35	1.37E-02	0.39	4.64E-03	1.22	2.52E-01	− 0.04	2.60E-01

Results shown are regression effect sizes and p-values for

^*HbA₂ adjusted for age and sex

^**HbA₂ adjusted for age, sex, and HbF

^***HbF adjusted for age and sex

⁺mediation analysis.

Italicized values in the mediation analysis are for an analysis where either no statistically significant adjusted effect on HbA₂ or no statistically significant association with HbF was observed for that SNP. These values should not be interpreted for mediation, but are provided for completeness.

Genotyping

The DNA from the CSSCD, Walk-PHaSST, PUSH and Hong Kong samples were genotyped using the Illumina Human610-Quad array and processed and analyzed as described.[17] We used the genome-wide identity by descent analysis in PLINK to discover unknown relatedness. Pairs with identity by descent measurements greater than 0.2 were deemed to be related subjects and only one subject was included in the analysis. We analyzed only common SNPS (MAF>0.05), from individuals with call rates of at least 98%.

Phenotype

HbA₂ was measured by DEAE Cellulose column chromatography [18, 19] (CSSCD) or high performance liquid chromatography (Walk-PHaSST, PUSH, β-thalassemia cohort). HbF was measured by alkali denaturation [20] (CSSCD) and HPLC (Walk-PHaSST, PUSH, β-thalassemia cohort). Measurements taken when a subject was under 5 years old were discarded as HbA₂ values may not be stable before this age. HbA₂ values of below 1.4 or above 7.9 were discarded in the sickle cell anemia cohorts, as values outside of these bounds are considered to be biologically unlikely and due to instrumentation or recording errors.

The CSSCD and Walk-PHaSST were both longitudinal studies, and some subjects had multiple HbA₂ and HbF measurements and median HbA₂, HbF and age measurements were used. In the CSSCD HbA₂ and HbF were measured on the same blood sample.

Statistical analysis and genome-wide association study

Data are represented as mean and standard deviation or median and range. Relation between HbA₂ and MCV was tested using linear regression, and the effect of α thalassemia was tested using an interaction between MCV and α thalassemia. Association between HbA₂ and each SNP was tested using an additive genetic model in PLINK, [21] and a normal distribution for HbA₂. The normal distribution assumption is shown by the QQ-plot (Figure S1). Two models were conducted, (1) adjusted for age and sex; (2) adjusted for age, sex, and HbF. SNPs that reached a significant association with HbA₂ with p<1E-5 in the CSSCD samples were assessed in the remaining cohorts. At this significance level, if no associations were present and all SNPs were independent, we would expect to observe 5.5 false positives. The presence of gene deletion α thalassemia was adjusted for in the CSSCD cohort. A meta-analyses using Metal software assessed all studies simultaneously.[22]

For replication of the 9 available SNPs from the 13 previously found associated with HbA₂ in normal individuals[2], we used a p-value threshold of p<0.05/9. To further examine previously reported associations, we conducted a mediation analysis to determine if there was mediation in the SNP/HbA₂ association by HbF.[23] Mediation was assessed by performing a series of regressions adjusted for age and sex:

$${\mathrm{HbA}}_2={\mathrm{i}}_1+\mathrm{c}\left(\mathrm{SNP}\right)+{\mathrm{e}}_1$$ (1)

$${\mathrm{HbA}}_2={\mathrm{i}}_2+{\mathrm{c}}^{\prime }\left(\mathrm{SNP}\right)+\mathrm{b}\left(\mathrm{HbF}\right)+{\mathrm{e}}_2$$ (2)

$$\mathrm{HbF}={\mathrm{I}}_3+\mathrm{a}\left(\mathrm{SNP}\right)+{\mathrm{e}}_3$$ (3)

If, for a SNP, c and a are significantly different from zero, these regressions are then used to evaluate the effect of SNP on HbA₂ through HbF by (c-c’) (Figure S2).[23] Values shown in Table 1 and Supplement Table 3 correspond to hypothesis tests of whether this mediated effect is nonzero.

Results

Patient Characteristics

CSSCD and PUSH patients were younger than Walk-PHaSST (Table S1).[24] The distribution of HbA ₂ in all cohorts is shown in Figure S3 A-D. The mean value of HbA ₂ was ~ 3% In the CSSCD, 3.8% in Walk-PHaSST and 3.5% in PUSH reflecting higher HbA₂ measurement by HPLC in the presence of HbS. The Hong Kong cohort with β-thalassemia trait had a mean of 5.4%. In this cohort, HbA₂ levels are also affected by the nature of the β-thalassemia mutation that also is associated with the numbers of F-cells.[16]

HbA₂ and HbF were correlated (r=-0.20, nonzero with p=3.7E-24; Figure 1) as were HbA₂ and MCV (r=0.58, p=1.8E-137) and to a lesser extent HbF and MCV (r=0.14, p=1.7E-08). The relationship between HbA₂ and MCV remained significant after adjusting for HbF (p<2E-16). α Thalassemia was associated with a reduction of MCV; (Figure S4). Increased HbA₂ in sickle cell anemia-α thalassemia is a result α/δ-globin dimerizing more readily than α/β^S-globin.[18] α Thalassemia modified the relationship between HbA₂ and MCV: without α thalassemia, a regression of HbA₂ on MCV yielded a slope of -0.028; with α thalassemia, the slope was -0.038 (a -33% change, p= 0.0339).(Table S2.)

Figure 1.

Association between HbA₂ and HbF. HbA₂ and HbF are correlated (r=-0.20; nonzero with p=3.7E-24).

Replication of association in northern Europeans

In northern Europeans the HSB1L-MYB intergenic region on chromosome 6q and the HBB cluster on chromosome 11p were strongly associated with HbA₂.[2] Of 13 reported associations, 9 SNPs were available for our analysis (Table 1). When adjusted for age and sex, nominal replication (p<0.05) was obtained for 4 SNPs in the HSB1L-MYB intergenic region (bold font in Table 1), and 1 SNP in the HBB gene cluster. Considering only SNPs with p<0.05 in the analysis adjusted for age, sex and HbF, only rs9494145 in HSB1L-MYB (p=4.5E-02), rs11036212 in HBB (p=4.5E-02), and rs11036364 in HBB (p=4.64E-03) were replicated. As reported, R² between 0.01 and 0.02 for their top SNPs, we have power between 0.70 and 0.94 to nominally replicate (p=0.05) these results if such an association exists. For these SNPs in the CSSCD, we obtained R² between 0.00017 and 0.0067.

The association between HbA₂ and the SNPs identified in [2] were tested for mediation by HbF in CSSCD samples. Only rs9494145 in chromosome 6q was associated with HbA₂ and this effect was mediated by HbF (p=6.33E-05). The effect of rs11036212 in chromosome 11p was not mediated by HbF. This is consistent with Menzel's results the results[2] as their associations in chromosome 11 remained strong even after adjustment for HbF.

GWAS results in CSSCD

GWAS in the CSSCD are shown in Figure 2, Table 2 and Supplement Tables 3 and 4. Rs6038123 with the smallest p-value adjusted for age, sex, and HbF (p=1.37E-06) (20p12.3; position 5, 291,506), is located downstream of UBE2D3P1 (ubiquitin-conjugating enzyme E2D 3 pseudogene 1). This SNP also had the largest partial R² (0.039) for any SNPs analyzed in the CSSCD samples. Uda et al found an association of rs6037828 (20p13; position 508,365 ) with HbF in Sardinians with β thalassemia.[25] In a meta-analysis of more than 2000 patients with sickle cell anemia we found no association of any SNP on chromosome 20 with HbF at p<1E-5.[26] The next-strongest association, adjusted for HbF, is at rs822278, located in the intergenic between MRPL49P2 (mitochondrial ribosomal protein L49 pseudogene 2) and FGF20 (fibroblast growth factor 20) on chromosome 8 (p=1.4E-06). These genes have not been implicated in hematological traits, and replication is required to ensure they are not false positives.

Figure 2.

Genome-wide association results for CSSCD adjusted for age, sex, and HbF. Left Panel: Manhattan plots of –log10(p-values) (y-axis) versus SNP positions (x-axis). No SNPs achieve genome-wide significance, but we find suggestive results on chromosomes 2, 6, and 11. Right Panel: QQ-plots for CSSCD GWAS results.

Table 2.

Top associations between SNPs and HbA₂ in CSSCD GWAS.

					Adjusted for age and sex		Adjusted for age, sex, and HbF		Mediation by HbF
CHR	SNP	BP	MAF	A1	Effect	P	Effect	P	P
2	rs766432	60573474	0.29	C	−0.195	8.10E-08	−0.1326	4.98E-04	3.50E-05
2	rs10195871	60574093	0.33	A	−0.166	1.69E-06	−0.112	1.64E-03	2.35E-05
2	rs17020489	81675692	0.3	A	0.1787	8.00E-07	0.1668	2.20E-06	1.73E-01
2	rs17020632	81765823	0.31	C	0.1667	4.17E-06	0.161	4.75E-06	5.13E-01
2	rs1113932	192760598	0.06	G	0.2881	2.00E-05	0.2895	9.95E-06	9.31E-01
8	rs822278	16536220	0.47	G	0.1613	8.89E-07	0.1539	1.40E-06	3.49E-01
8	rs1425912	54241746	0.36	G	0.1572	7.65E-06	0.1532	7.09E-06	6.37E-01
10	rs10887055	123789677	0.26	G	0.1746	2.51E-06	0.1523	2.70E-05	1.55E-02
11	rs11021763	11273929	0.37	G	0.1474	2.39E-05	0.1502	9.09E-06	7.45E-01
13	rs1536807	40267790	0.44	A	−0.1514	8.11E-06	−0.1371	3.39E-05	8.19E-02
13	rs2940695	107365156	0.25	A	0.175	5.24E-06	0.1665	8.19E-06	3.61E-01
14	rs2296274	60986931	0.15	A	−0.2204	3.49E-06	−0.2063	8.00E-06	2.19E-01
16	rs11150150	77897032	0.34	G	−0.1627	4.21E-06	−0.1501	1.31E-05	1.39E-01
20	rs6038123	5220152	0.2	A	−0.2138	4.69E-06	−0.2193	1.37E-06	6.19E-01

SNPs significant at p<1E-05 in at least one analysis of CSSCD.

Replication in secondary cohorts was attempted for all CSSCD variants with p<1E-5 (Table S3, age and sex adjusted, and Table S4, age, sex, and HbF adjusted). Nominal replication (p<0.05) was achieved for rs766432 and rs10195871 in BCL11A. Both SNPs are associated with HbF levels.[17, 27-33] Rs766432 achieved genome-wide significance in meta-analysis (adjusted for age and sex only), with p=3.03E-10. All SNPs in Tables S3 and S4 reached statistical significance after Bonferroni correction (p < 0.05/14=0.004), and several SNPs had also consistent effects in the different studies. Mediation analysis suggested that the association of HbA₂ with BCL11A is at least partially mediated by HbF.

Discussion

Normal newborns have less than 0.3% HbA₂.[34] We studied the genetic regulation of HbA₂ in sickle cell anemia and confirmed in part associations previously reported in normal northern Europeans.[2] A mediation analysis suggested, as did the studies of Menzel et al, that SNPs associated with HbA₂ in the HBS1L-MYB locus on chromosome 6q affect HbA₂ levels partially through their effects on HbF. An association of SNPs in BCL11A with HbA₂, is also mediated by the effects of this gene on HbF expression. The enhancer of MYB within HMIP-2 of the HBS1LMYB locus [8, 35-37] is likely to modulate HbF by dual effects on erythropoiesis and the activation of other genes like KLF1 that modulate HbF expression.[37] Low MYB levels accelerate erythroid differentiation leading to release of early progenitors synthesizing predominantly HbF and also directly affect γ-globin gene expression.[38, 39]

The association of SNPs within the HBB cluster and HbA₂ levels was not mediated by HbF. SNPs previously associated with HbA₂ [2], and the SNP we replicated (rs11036212), are in the 3’ hypersensitive site (HS; downstream) of the HBB cluster, whereas SNPs associated with HbF are 5’ (upstream) of HBB. We could not replicate the associations found in Sardinians with β-thalassemia trait where the associations were in the 5’ β-globin gene locus control region.[26] Our results explain only a small amount of the variance of HbA₂ that is likely to be affected by the complex regulation of globin gene expression.[40]

HbF is the major modulator of the phenotype of sickle cell anemia, but HbA₂ has similar effects on HbS polymerization.[9, 41] Whether HBD expression can be increased, perhaps by targeting the mutated CACC binding site for KLF1 in the HBD promoter[7], is unknown. Other differences in HBD might thwart this approach.[42-49] Pancellular expression of HbA₂ predicts that increased HbA₂ levels could be especially beneficial.[50]

Gene expression within the HBB gene cluster is partially a function of competition among proximal gene promoters for transcription factors and the β-globin locus control region.[51] This is exemplified by increases in HbF and HbA₂ when the promoter of the β-globin gene is deleted or mutated.[52-55] Substantial increases in HbA₂ and HbF was found in HbS-β⁰ thalassemia when the β-globin gene promoter was deleted, and this was associated with mild disease.[56]

HbF was measured by two different methods our cohorts examined; within the range of HbF observed, both give similar results. Nevertheless, this could be a confounder.[57] HbA₂ was also measured by differed means. When done by HPLC (Walk-PHaSST and PUSH trials) HbA₂ levels are increased due to the co-elution of some HbS adducts.[58]

In sickle cell anemia, levels of HbA₂ appear to be modulated by genes that directly and indirectly effect HBG expression and perhaps also by regulation within the β-globin gene cluster. It remains to be seen if these observations can be exploited for therapeutic purposes.